Anupam0202/mavunoguard-ai

MavunoGuard is a mobile-first, Gemini-powered post-harvest decision-support prototype for maize farmers, cooperatives, buyers, and extension officers in Eldoret and Kenya’s North Rift. A farmer adds one grain photo plus storage context; the app returns an explainable risk score, bilingual safe-action sequence, rescue-value scenario, test referral. MavunoGuard AI See risk. Save grain. Prove quality. Mobile-first post-harvest decision support for maize farmers, cooperatives, buyers, and extension teams. --- ## Overview MavunoGuard turns a grain photo and a few field observations into an explainable post-harvest risk assessment. It combines Gemini multimodal analysis with a transparent rule engine to produce: - Visible-condition findings - A traceable risk score and its contributing factors - English and Kiswahili action plans - A certified-testing referral pathway - An illustrative grain-value protection scenario - A printable batch passport - An M-Pesa test-booking simulation MavunoGuard is designed to help users decide what to do next with uncertain grain—not to replace laboratory testing. > [!IMPORTANT] > MavunoGuard does **not** detect, quantify, confirm, or rule out aflatoxin or certify food/feed safety. Suspect batches require an approved rapid or laboratory test. ## Product flow ```text Capture evidence │ ▼ Photo + moisture + storage + weather + observations │ ├───────────────┐ ▼ ▼ Gemini review Deterministic rules │ │ └───────┬───────┘ ▼ Explainable risk assessment │ ┌───────┼────────┐ ▼ ▼ ▼ Safe plan Test path Value scenario │ ▼ Batch passport ``` ## Features ### Multimodal assessment Upload or capture a maize-grain image and combine it with moisture, storage method, storage duration, recent weather, odour, mould, insect damage, and wet-bagging observations. ### Structured Gemini output The server calls Gemini using a constrained JSON schema so the interface receives predictable findings, actions, confidence, and uncertainty fields. ### Explainable risk engine The final risk score is not generated from an opaque prompt alone. A deterministic engine applies visible, reviewable weights to the supplied field conditions and combines them with the visual-condition category. ### Safety-first recommendations System instructions prevent t …